AI Succeeds at Opportunistic Screening for Osteoporosis
Identification of low bone density enabled through secondary analysis of CT scans
A deep learning (DL) model can identify signs of osteoporosis in CT images obtained for other reasons, according to research conducted by a team in New York and Germany. Their study, published in Radiology, advances the feasibility of using opportunistic screening to address the underdiagnosis and treatment of osteoporosis.
More than two million osteoporosis-related fractures occur annually in the United States, resulting in high levels of morbidity and mortality. “We have treatments for osteoporosis, so we can do things to prevent fractures, but a lot of people don’t get tested for osteoporosis,” said study author, Miriam Bredella, MD, MBA, from the Department of Radiology at NYU Langone Health and the NYU Grossman School of Medicine in New York City.
Dr. Bredella and colleagues reported that just one in five eligible patients undergo osteoporosis screening. They explored whether AI could reexamine imaging to improve that rate. “It’s so much better to prevent fractures rather than dealing with the consequences,” Dr. Bredella said.
AI Analyzed More Than a Half Million Images
Working with their PACS vendor, the researchers fine-tuned an existing algorithm using NYU data and applied it to 538,946 CT exams of 283,499 patients in NYU’s clinical archive (2003–2024) to identify low bone density in the middle and lower spine.
The team also sought to establish normative values and diagnostic thresholds for the CT images, using World Health Organization data on osteoporosis and osteopenia prevalence to set these standards.
“We came up with thresholds for each vertebral body, starting from the first thoracic vertebra (T1) all the way to the lower spine,” Dr. Bredella said.
To verify the analysis, three experienced NYU radiologists manually reviewed 1,496 randomly selected studies, assessing the AI’s placements of regions of interest on the scans and identification of the vertebral bodies. They agreed with more than 99% of the findings.
Along with testing their DL model’s capabilities, the researchers sought to develop a statistical method to adjust for the use of different CT scanners and scanning protocols to generate images. “That’s quite genius, because we cannot calibrate every scanner with a phantom,” said Georg Feuerriegel, MD, of the Department of Radiology and Biomedical Imaging at the University of California, San Francisco, coauthor of a related editorial. “They used a statistical harmonization approach to correct attenuation values across different tube voltages and scanner models.”
Applying AI Innovation to Clinical Workflow
The researchers suggest that AI-enabled opportunistic screening could identify more patients at risk for osteoporosis and would prompt referral for dual-energy X-ray absorptiometry (DXA) imaging, enabling earlier treatment and helping prevent fractures and related complications.
The technology to support opportunistic screening is already available with commercial tools and internally developed models offering potential pathways for implementation. Additionally, Dr. Bredella said radiology departments could develop their own AI tools for opportunistic screening.
“We’re still searching for really useful AI tools in clinical routine,” Dr. Feuerriegel noted. “This could be one of the clinical applications that could really improve the workflow for the radiologists.”
However, in his editorial, Dr. Feuerriegel cautions that opportunistic screening must be paired with clinical processes to produce potential results. Consequently, studies will need to determine what combination of screening and clinical follow-up, if any, produce improved patient outcomes.
Dr. Bredella emphasized that integrating follow-up into clinical workflows is essential. At NYU Langone Health, when such screening shows low bone density, the patient is automatically added to their electronic medical system work list of cases needing follow-up.
“A dedicated bone health team follows up every patient identified as having bone loss on CT,” she said. “They contact patients directly, arrange DXA testing, and coordinate referrals to endocrinology or rheumatology when needed for treatment so no patient falls through the cracks.”
Other Potential Applications
Opportunistic screening using AI has other potential applications. NYU Langone Health now is using it to quantify calcification in breast arteries that are imaged in mammograms. “You could capture patients at risk of cardiovascular disease along with breast cancer,” Dr. Bredella explained.
Dr. Feuerriegel also sees potential for assessing side effects of treatments for certain conditions. “Cancer, inflammation…all can have an effect on bone density,” he observed. “When a patient undergoes imaging to assess their condition and their treatment’s effectiveness, the scans potentially also could be used to measure the effects on bones.”
Both doctors think AI-driven opportunistic CT screening has potential to reduce the volume of DXA imaging. “DXA will not disappear,” Dr. Feuerriegel said. “But opportunistic CT screening could substantially reduce the need for dedicated screening examinations.”
For now, however, patients identified through opportunistic CT screening typically undergo additional evaluation, including DXA testing, as part of clinical follow-up.
Dr. Bredella said broader adoption will depend on insurers accepting CT-based detection of low bone density, rather than requiring confirmation with DXA to approve the need for osteoporosis medication. Future prospective studies may further evaluate the approach and its impact on patient outcomes.
“There is so much information on these studies that we’re just not using, that we’re throwing away, even though it can be incredibly helpful for prevention,” Dr. Bredella said.
For More Information
Access the Radiology study, “Deep Learning-based Opportunistic CT Osteoporosis Screening and the Establishment of Normative Values,” and related editorial, “Using AI for Opportunistic Osteoporosis Screening at CT- Ready for Routine Care.”
Read previous RSNA News stories on opportunistic screening: